Combinatorial optimisation problems (COPs) have been at the origin of the design of many optimal and heuristic solution frameworks such as branch-and-bound algorithms, branch-and-cut algorithms, classical local search methods, metaheuristics, and hyperheuristics. This thesis proposes a refined generic and parametrised infeasible local search (GPILS) algorithm for solving COPs and customises it to solve the traveling salesman problem (TSP), for illustration purposes. In addition, a rule-based heuristic is proposed to initialise infeasible local search, referred to as the parameterised infeasible heuristic (PIH), which allows the analyst to have some control over the features of the infeasible solution he/she might want to start the infeasible search with. A recursive infeasible neighbourhood search (RINS) as well as a generic patching procedure to search the infeasible space are also proposed. These procedures are designed in a generic manner, so they can be adapted to any choice of parameters of the GPILS, where the set of parameters, in fact for simplicity, refers to set of parameters, components, criteria and rules. Furthermore, a hyperheuristic framework is proposed for optimizing the parameters of GPILS referred to as HH-GPILS. Experiments have been run for both sequential (i.e. simulated annealing, variable neighbourhood search, and tabu search) and parallel hyperheuristics (i.e., genetic algorithms / GAs) to empirically assess the performance of the proposed HH-GPILS in solving TSP using instances from the TSPLIB. Empirical results suggest that HH-GPILS delivers an outstanding performance. Finally, an offline learning mechanism is proposed as a seeding technique to improve the performance and speed of the proposed parallel HH-GPILS. The proposed offline learning mechanism makes use of a knowledge-base to keep track of the best performing chromosomes and their scores. Empirical results suggest that this learning mechanism is a promising technique to initialise the GA's population.

In this paper we study a complex real-world workforce scheduling problem. We propose a method of splitting the problem into smaller parts and solving each part using exhaustive search. These smaller parts comprise a combination of choosing a method to select a task to be scheduled and a method to allocate resources, including time, to the selected task. We use reduced Variable Neighbourhood Search (rVNS) and hyperheuristic approaches to decide which sub problems to tackle. The resulting methods are compared to local search and Genetic Algorithm approaches. Parallelisation is used to perform nearly one CPU-year of experiments. The results show that the new methods can produce results fitter than the Genetic Algorithm in less time and that they are far superior to any of their component techniques. The method used to split up the problem is generalisable and could be applied to a wide range of optimisation problems.

Bayesian network (BN) structure learning from data has been an active research area in the machine learning field in recent decades. Much of the research has considered BN structure learning as an optimization problem. However, the finding of optimal BN from data is NP-hard. This fact has driven the use of heuristic algorithms for solving this kind of problem. Amajor recent focus in BN structure learning is on search and score algorithms. In these algorithms, a scoring function is introduced and a heuristic search algorithm is used to evaluate each network with respect to the training data. The optimal network is produced according to the best score evaluated. This thesis investigates a range of search and score algorithms to understand the relationship between technique performance and structure features of the problems. The main contributions of this thesis include (a) Two novel Ant Colony Optimization based search and score algorithms for BN structure learning; (b) Node juxtaposition distribution for studying the relationship between the best node ordering and the optimal BN structure; (c) Fitness landscape analysis for investigating the di erent performances of both chain score function and the CH score function; (d) A classifier method is constructed by utilizing receiver operating characteristic curve with the results on fitness landscape analysis; and finally (e) a selective o -line hyperheuristic algorithm is built for unseen BN structure learning with search and score algorithms. In this thesis, we also construct a new algorithm for producing BN benchmark structures and apply our novel approaches to a range of benchmark problems and real world problem.

Le succès dans l'utilisation de méthodes exactes d’optimisation combinatoire pour des problèmes de grande taille est encore limité à certains problèmes ou à des classes spécifiques d'instances de problèmes. Une approche alternative consiste soit à utiliser des métaheuristiques ou des matheuristiques qui reposent en partie sur des méthodes exactes. Dans le contexte de l'optimisation combinatoire, nous nous intéressons des heuristiques permettant de choisir les heuristiques appliquées au problème traité. Dans cette thèse, nous nous concentrons sur l'optimisation à l’aide d’hyperheuristiques pour des problèmes logistiques. Nous proposons un cadre hyperheuristique qui effectue une recherche dans l'espace des algorithmes heuristiques et apprend comment changer l'heuristique courante systématiquement tout au long du processus de telle sorte qu'une bonne séquence d'heuristiques permet d’obtenir des solutions de haute qualité. Nous étudions plus particulièrement deux problèmes en logistique pour lesquels nous proposons des HHs: un problème de planification d’interventions sur des puits de forage et un problème conjoint de localisation de hubs et de routage. Ensuite, nous comparons les performances de plusieurs HH décrites dans la littérature pour le second problème abordé reposant sur différentes méthodes de sélection heuristique telles que la sélection aléatoire, la fonction de choix, une approche de Q-Learning et un algorithme de colonie de fourmis. Les résultats numériques prouvent l'efficacité de HHs pour les deux problèmes traités, et la pertinence d'inclure l'information venant d’une relaxation de Lagrangienne pour le deuxième problème. / Success in using exact methods for large scale combinatorial optimization is still limited to certain problems or to specific classes of instances of problems. The alternative way is either using metaheuristics or matheuristics that rely on exact methods in some ways. In the context of combinatorial optimization, we are interested in heuristics to choose heuristics invoked to solve the addressed problem. In this thesis, we focus on hyperheuristic optimization in logistic problems. We focus on proposing a hyperheuristic framework that carries out a search in the space of heuristic algorithms and learns how to change the incumbent heuristic in a systematic way along the process in such a way that a good sequence of heuristics produces high quality solutions. We propose HHs for two problems in logistics: the workover rig scheduling problem and the hub location routing problem. Then, we compare the performances of several HHs described in the literature for the latter problem, which embed different heuristic selection methods such as a random selection, a choice function, a Q-Learning approach, and an ant colony based algorithm. The computational results prove the efficiency of HHs for the two problems in hand, and the relevance of including Lagrangian relaxation information for the second problem.